I'd like to request that the labs prioritize spending their unlimited premium token budgets figuring out how to ship updates that don't require restarting each agent desktop app every day.
1
6 comments
Jamon
@jamonholmgren
On my code base, everything Codex seems to work better than everything Claude.
Very interesting results.
JJamonXView on XHere's the feature tier list. Again, Sol and Astra did great, Luna punched above its weight as did Sonnet. Haiku did very poorly. Opus not S-tier, but pretty good ... still made many more mistakes than Astra and Sol did.